|
Abstract
In traditional clustering problems, a distance measure is
usually defined over the whole set of attributes, and the goal is to
minimize intra-cluster object distance and maximize inter-cluster object
distance. In recent years, as the datasets involved in clustering
problems contain more and more attributes, it has been noticed that such
kind of distance measures may not be appropriate. Each cluster may
relate to only a subset of attributes, so considering the values of all
attributes may add noise to the distance calculation. In this seminar,
some different approaches to tackling the problem will be described, and
a special focus will be put on the projective clustering approach.
Read the Presentation
Slides...
Referred Papers
|